Upload streamlit_app.py

streamlit_app.py

@@ -1,13 +1,6 @@
-# streamlit_app.py — LamboVision · AI Car Studio (frontend) v0.6.3
-# Previews: 9:16 cards (Input / Packshot / Padded / Scene / Video)
-# Configurator: model, body color, finish, wheel style, caliper color + Extra prompt
-# Endpoints used:
-#   POST /v1/tryon/chain   (recommended full pipeline)
-#   POST /v1/video/from-image (on-demand video)
-#
-# Env:
-#   AI_LIGHTBOX_API   -> http://host:port (default: http://127.0.0.1:8000)
-#   AI_LIGHTBOX_TOKEN -> Bearer (optional)
+# streamlit_app.py — LamboVision · AI Car Studio (uniform cards) v0.6.4
+# All cards have identical size; images/videos are fit inside the frame (no overflow).
+# Default aspect is 3:4 (less tall), but user can switch to 9:16 or 1:1.
 
 import os, io, base64, zipfile, requests, streamlit as st
 from PIL import Image
@@ -30,7 +23,8 @@ HEADERS = {"Authorization": f"Bearer {HF_TOKEN}"} if HF_TOKEN else {}
 VIDEO_MAX_PX_DEFAULT = int(os.getenv("VIDEO_MAX_PX", "720"))
 VIDEO_RES_OPTIONS = ["512P", "768P"]
 VIDEO_DUR_OPTIONS = ["6", "10"]
-ASPECT_OPTIONS = ["9:16", "3:4", "1:1"]
+# Default 3:4 to avoid very tall cards; user can switch
+ASPECT_OPTIONS = ["3:4", "9:16", "1:1"]
 
 # ==== Session ====
 if "results" not in st.session_state:
@@ -65,16 +59,9 @@ def post_chain(data: dict, files_payload):
     r.raise_for_status()
     return r.json()
 
-def post_video(image_url: str, duration="6", resolution="768P", prompt_optimizer=False, preview_aspect="9:16", prompt=None):
-    payload = {
-        "image_url": image_url,
-        "duration": duration,
-        "resolution": resolution,
-        "prompt_optimizer": "true" if prompt_optimizer else "false",
-        "preview_aspect": preview_aspect,
-    }
-    if prompt is not None:
-        payload["prompt"] = prompt
+def post_video(image_url: str, duration="6", resolution="768P", preview_aspect="3:4", prompt=None):
+    payload = {"image_url": image_url, "duration": duration, "resolution": resolution, "preview_aspect": preview_aspect}
+    if prompt is not None: payload["prompt"] = prompt
     r = requests.post(f"{API_BASE}/v1/video/from-image", data=payload, headers=HEADERS, timeout=600)
     r.raise_for_status()
     return r.json()
@@ -99,8 +86,7 @@ def _sniff_mime_from_bytes(b: bytes) -> str:
         return "image/jpeg"
 
 def image_bytes_to_data_url(b: bytes, mime: str | None = None) -> str:
-    if not mime:
-        mime = _sniff_mime_from_bytes(b)
+    if not mime: mime = _sniff_mime_from_bytes(b)
     enc = base64.b64encode(b).decode("ascii")
     return f"data:{mime};base64,{enc}"
 
@@ -117,8 +103,7 @@ def infer_ext(data: bytes) -> str:
         fmt = (im.format or "JPEG").lower()
         return "." + {"jpeg":"jpg","jpg":"jpg","png":"png","webp":"webp"}.get(fmt,"jpg")
     except Exception:
-        if data[:4] == b"\x00\x00\x00\x18" or data[4:8] == b"ftyp":
-            return ".mp4"
+        if data[:4] == b"\x00\x00\x00\x18" or data[4:8] == b"ftyp": return ".mp4"
         return ".bin"
 
 def make_zip(named_bytes: list[tuple[str, bytes]]) -> bytes:
@@ -128,12 +113,12 @@ def make_zip(named_bytes: list[tuple[str, bytes]]) -> bytes:
             if b: z.writestr(fname, b)
     buf.seek(0); return buf.read()
 
-# ==== Preview helpers (uniform 9:16 cards) ====
+# ==== Preview helpers ====
 def _parse_aspect_str(s: str):
     try:
         a, b = s.split(":"); return max(1, int(a)), max(1, int(b))
     except Exception:
-        return (9, 16)
+        return (3, 4)
 
 def inject_base_css():
     st.markdown("""
@@ -141,39 +126,61 @@ def inject_base_css():
       .stApp { background: #0e0f12; }
       header[data-testid="stHeader"] { background: transparent; }
       .block-container { padding-top: 1.1rem; }
+
       h1, .stMarkdown h1 { font-size: 1.85rem; line-height: 1.25; margin-bottom: 0.1rem; }
       .stMarkdown p { margin-top: 0.15rem; }
+
       label, .stTextInput label, .stSelectbox label, .stNumberInput label, .stFileUploader label,
       .stCheckbox label, .stSlider label, .stRadio label {
-        color: #ff8a33 !important;
-        font-weight: 600 !important;
+        color: #ff8a33 !important; font-weight: 600 !important;
       }
+
       .stButton > button[kind="primary"]{
         background: #ff7a1a !important; border: 0 !important; color: #111 !important;
       }
       .stButton > button:hover { filter: brightness(1.05); }
+
       .stAlert { border-radius: 10px; }
+
       .lv-card-title { font-weight: 700; font-size: 1.05rem; margin: 2px 0 8px 0; color: #eaecef; }
+
+      /* UNIFORM FRAME: one box size for all media; content never overflows */
       .lv-frame {
-        position: relative; width: 100%;
-        border-radius: 12px; overflow: hidden;
+        position: relative;
+        width: 100%;
+        aspect-ratio: 3 / 4;    /* default; overridden by inject_preview_css */
+        border-radius: 12px;
+        overflow: hidden;
         background: #111419;
-        display:flex; align-items:center; justify-content:center;
+        display: flex; align-items: center; justify-content: center;
         box-shadow: 0 0 0 1px rgba(255,122,26,0.15) inset;
       }
-      .lv-frame img { width:100%; height:100%; display:block; }
-      .lv-video { display:flex; justify-content:center; }
-      .lv-video video { border-radius:8px; max-width: 540px; width:100%; height:auto; }
-      .stCaption, .st-emotion-cache-1r6slb0 { color: #ffb07a !important; }
+
+      /* CONTAIN for both IMG and VIDEO (no cropping, no spill) */
+      .lv-frame > img,
+      .lv-frame > video,
+      .lv-frame > .lv-media {
+        width: 100%;
+        height: 100%;
+        object-fit: contain !important;
+        display: block;
+        background: transparent;
+      }
+
+      /* reset inner element spacing to avoid shifting card height */
+      .lv-frame * { margin: 0 !important; padding: 0 !important; border: 0 !important; }
     </style>
     """, unsafe_allow_html=True)
 
 def inject_preview_css(aspect: str, fit_mode="contain"):
-    aw, ah = _parse_aspect_str(aspect)
+    try:
+        a, b = [int(x) for x in aspect.split(":")]
+    except Exception:
+        a, b = 3, 4
     st.markdown(f"""
     <style>
-      .lv-frame {{ aspect-ratio: {aw} / {ah}; }}
-      .lv-frame img {{ object-fit:{fit_mode}; }}
+      .lv-frame {{ aspect-ratio: {a} / {b}; }}
+      .lv-frame > img, .lv-frame > video, .lv-frame > .lv-media {{ object-fit: {fit_mode}; }}
     </style>
     """, unsafe_allow_html=True)
 
@@ -202,21 +209,21 @@ def make_small_jpeg_from_native(native: bytes, target_long_edge: int = 1280, bg_
 
 def make_preview_frame(img_bytes: bytes, aspect: str, long_edge: int, ratio: float, bg_rgb=(255,255,255)) -> bytes:
     im = Image.open(io.BytesIO(img_bytes)).convert("RGBA")
-    aw, ah = _parse_aspect_str(aspect)
-    if aw >= ah:
-        W = max(256, min(8192, int(long_edge))); H = max(1, int(W * ah / aw))
+    a, b = _parse_aspect_str(aspect)
+    if a >= b:
+        W = max(256, min(8192, int(long_edge))); H = max(1, int(W * b / a))
     else:
-        H = max(256, min(8192, int(long_edge))); W = max(1, int(H * aw / ah))
+        H = max(256, min(8192, int(long_edge))); W = max(1, int(H * a / b))
     ratio = max(0.10, min(0.99, float(ratio)))
-    target = int(min(W, H) * ratio if aw == ah else (H * ratio if aw < ah else W * ratio))
+    target = int(min(W, H) * ratio if a == b else (H * ratio if a < b else W * ratio))
     w, h = im.size
-    if w >= h: new_w, new_h = target, max(1, int(h * (target / max(1, w))))
-    else:      new_h, new_w = target, max(1, int(w * (target / max(1, h))))
+    if w >= h: new_w, new_h = target, max(1, int(h * (new_w / max(1, w))))
+    else:      new_h, new_w = target, max(1, int(w * (new_h / max(1, h))))
     im_resized = im.resize((new_w, new_h), Image.LANCZOS)
     canvas = Image.new("RGBA", (W, H), (*bg_rgb, 255))
     off = ((W - new_w)//2, (H - new_h)//2); canvas.paste(im_resized, off, im_resized)
     out = io.BytesIO(); canvas.convert("RGB").save(out, format="JPEG", quality=90, subsampling=1)
-    return out.getvalue()
+    return out.read()
 
 # ==== Local demo files (optional) ====
 def load_local_demo():
@@ -289,15 +296,15 @@ with colF:
 
 colT1, colT2, colT3, colT4 = st.columns([1.0, 1.0, 1.0, 1.2])
 with colT1:
-    preview_aspect = st.selectbox("Preview Frame Ratio", ASPECT_OPTIONS, index=0, key="preview_aspect")
+    preview_aspect = st.selectbox("Preview Frame Ratio", ASPECT_OPTIONS, index=0, key="preview_aspect")  # default 3:4
 with colT2:
-    preview_long_edge = st.number_input("Preview Long Edge (px)", min_value=256, max_value=8192, value=1920, step=64, key="preview_long_edge")
+    preview_long_edge = st.number_input("Preview Long Edge (px)", min_value=256, max_value=8192, value=1536, step=64, key="preview_long_edge")
 with colT3:
     fit_mode = st.selectbox("Preview Layout", ["fit","fill"], index=0, help="Affects previews only.", key="fit_mode")
 with colT4:
     st.info(f"Connection: {'Online' if ok else 'Offline'}", icon="🔌")
 
-# Inject CSS for aspect
+# Apply uniform frame CSS
 inject_preview_css(preview_aspect, fit_mode="contain" if fit_mode=="fit" else "cover")
 
 # ==== Build edit prompt (cars) ====
@@ -314,20 +321,10 @@ def build_edit_prompt(model, color, finish, wheel, caliper, extra):
         parts.append(extra.strip())
     return ", ".join(parts)
 
-# ==== Pre-run PREVIEWS (5 cards) ====
-st.subheader("Previews (9:16)")
+# ==== Pre-run PREVIEWS (5 uniform cards) ====
+st.subheader("Previews (Uniform Cards)")
 c_in, c_pack, c_pad, c_scene, c_video = st.columns(5)
 
-# Input bytes
-input_bytes = None
-if file_list:
-    try: input_bytes = file_list[0].getvalue()
-    except Exception: input_bytes = None
-elif image_url:
-    input_bytes = fetch_bytes(image_url)
-elif demo_files.get("input.jpg"):
-    input_bytes = demo_files["input.jpg"]
-
 def show_tile_card_display_download(
     col, title,
     display_url: str | None = None,
@@ -338,21 +335,22 @@ def show_tile_card_display_download(
     key_prefix: str = "",
 ):
     col.markdown(f'<div class="lv-card-title">{title}</div>', unsafe_allow_html=True)
-    if not display_url and not display_bytes:
-        col.markdown('<div class="lv-frame"></div>', unsafe_allow_html=True)
-        col.caption("—"); return None, None
 
+    # FRAME (always same size)
     if display_bytes is not None:
         src = image_bytes_to_data_url(display_bytes)
-        col.markdown(f'<div class="lv-frame"><img src="{src}"/></div>', unsafe_allow_html=True)
-    else:
+        col.markdown(f'<div class="lv-frame"><img class="lv-media" src="{src}"/></div>', unsafe_allow_html=True)
+    elif display_url:
         abs_url = _abs_backend_url(display_url)
-        col.markdown(f'<div class="lv-frame"><img src="{abs_url}"/></div>', unsafe_allow_html=True)
+        col.markdown(f'<div class="lv-frame"><img class="lv-media" src="{abs_url}"/></div>', unsafe_allow_html=True)
+    else:
+        col.markdown('<div class="lv-frame"></div>', unsafe_allow_html=True)
+        col.caption("—"); return None, None
 
+    # Native (caption + download)
     native = download_bytes or (fetch_bytes(download_url) if download_url else None)
     if not native:
         native = display_bytes
-
     if native:
         sz = image_size_from_bytes(native)
         if sz: col.caption(f"Real Res: {sz[0]}×{sz[1]} px")
@@ -366,8 +364,18 @@ def show_tile_card_display_download(
 
     return display_url or "data", native
 
+# Input bytes
+input_bytes = None
+if file_list:
+    try: input_bytes = file_list[0].getvalue()
+    except Exception: input_bytes = None
+elif image_url:
+    input_bytes = fetch_bytes(image_url)
+elif demo_files.get("input.jpg"):
+    input_bytes = demo_files["input.jpg"]
+
 show_tile_card_display_download(
-    c_in, "Sample Input preview",
+    c_in, "Sample Input",
     display_bytes=input_bytes if input_bytes else demo_files.get("input.jpg"),
     download_bytes=input_bytes if input_bytes else demo_files.get("input.jpg"),
     filename_stub="input", key_prefix="pre"
@@ -375,7 +383,7 @@ show_tile_card_display_download(
 
 pack_bytes_demo = demo_files.get("packshot.jpg") or input_bytes
 show_tile_card_display_download(
-    c_pack, "Sample Packshot (Clean/Configured)",
+    c_pack, "Sample Packshot (Configured)",
     display_bytes=pack_bytes_demo,
     download_bytes=pack_bytes_demo,
     filename_stub="packshot", key_prefix="pre"
@@ -390,7 +398,7 @@ if pack_bytes_demo:
     except Exception:
         padded_bytes = None
 show_tile_card_display_download(
-    c_pad, "Sample Padded (9:16 UI)",
+    c_pad, "Sample Padded (UI)",
     display_bytes=padded_bytes or pack_bytes_demo,
     download_bytes=padded_bytes or pack_bytes_demo,
     filename_stub="padded_ui", key_prefix="pre"
@@ -398,20 +406,20 @@ show_tile_card_display_download(
 
 scene_bytes_demo = demo_files.get("scene.jpg") or pack_bytes_demo
 show_tile_card_display_download(
-    c_scene, "Sample Scene Composite",
+    c_scene, "Sample Scene",
     display_bytes=scene_bytes_demo,
     download_bytes=scene_bytes_demo,
     filename_stub="scene_composite", key_prefix="pre"
 )
 
-c_video.markdown('<div class="lv-card-title">AI Generated Video Preview</div>', unsafe_allow_html=True)
+c_video.markdown('<div class="lv-card-title">Sample Video</div>', unsafe_allow_html=True)
 vid_demo = demo_files.get("tryon.mp4")
 if vid_demo:
     data_url = "data:video/mp4;base64," + base64.b64encode(vid_demo).decode("ascii")
     c_video.markdown(
         f"""
-        <div class="lv-frame" style="display:flex;align-items:center;justify-content:center;">
-          <video autoplay loop muted playsinline style="width:100%;height:100%;object-fit:cover;">
+        <div class="lv-frame">
+          <video class="lv-media" autoplay loop muted playsinline>
             <source src="{data_url}" type="video/mp4">
           </video>
         </div>
@@ -435,7 +443,19 @@ with colAct4:
 with colAct5:
     run = st.button("Run ▶", type="primary", use_container_width=True, key="run_btn")
 
-# ==== Run pipeline ====
+# ==== Pipeline ====
+def build_edit_prompt(model, color, finish, wheel, caliper, extra):
+    parts = [
+        f"Lamborghini {model}",
+        f"body color: {color}",
+        f"finish: {finish}",
+        f"wheel style: {wheel}",
+        f"brake calipers: {caliper}",
+        "catalog-clean alignment, keep true geometry, badges and proportions, realistic reflections"
+    ]
+    if extra and extra.strip(): parts.append(extra.strip())
+    return ", ".join(parts)
+
 if run:
     if not ok:
         st.error("Backend unreachable. Check API_BASE / TOKEN.")
@@ -443,48 +463,38 @@ if run:
         st.error("Provide at least one image (upload or URL).")
     else:
         try:
-            # Build prompts
             edit_prompt = build_edit_prompt(model_sel, body_color, finish, wheel, caliper, extra_prompt)
             scene_text = SCENE_PRESETS.get(scene_name, "premium studio scene")
             img_urls = image_url or ""
-
             files_payload = []
             if file_list:
                 for f in file_list:
                     files_payload.append(("files", (f.name, f.getvalue(), f.type or "image/jpeg")))
-
             data = {
                 "category": "auto",
                 "edit_prompt": edit_prompt,
                 "num_images": "1",
                 "image_urls": img_urls,
-                # scene pass
                 "do_scene_pass": "true" if do_scene_pass else "false",
                 "scene_prompt_text": scene_text,
-                # identity lock (backendside rules)
                 "identity_lock": "true" if identity_lock else "false",
-                # canvas/preview
                 "preview_aspect": preview_aspect,
                 "preview_long_edge": str(int(preview_long_edge)),
                 "preview_ratio": str(float(preview_ratio)),
-                # SR
                 "upscale": "true" if upscale else "false",
                 "upscale_factor": "2",
                 "upscale_stage": st.session_state.get("upscale_stage","both"),
-                # video
                 "to_video": "true" if to_video else "false",
-                "video_prompt": scene_text,  # same vibe for camera move
+                "video_prompt": scene_text,
                 "duration": st.session_state.get("duration_sel", VIDEO_DUR_OPTIONS[0]),
                 "resolution": st.session_state.get("resolution_sel", VIDEO_RES_OPTIONS[1]),
                 "prompt_optimizer": "false",
             }
-
             with st.spinner("LamboVision pipeline running…"):
                 out = post_chain(data, files_payload if files_payload else None)
 
             st.session_state["job_counter"] += 1
 
-            # Parse outputs
             pack_native_url = None; pack_display_jpeg = None
             try:
                 pack_native_url = (out["packshot_step"]["result"]["images_native"] or [{}])[0].get("url")
@@ -543,11 +553,11 @@ if run:
         except Exception as e:
             st.error(f"Error: {e}")
 
-# ==== Results ====
+# ==== Results (Uniform Cards) ====
 res = st.session_state.get("results")
 vbytes = None
 if res:
-    st.subheader("Results")
+    st.subheader("Results (Uniform Cards)")
     c_pack2, c_pad2, c_scene2, c_video2 = st.columns(4)
 
     show_tile_card_display_download(
@@ -559,7 +569,7 @@ if res:
     )
 
     show_tile_card_display_download(
-        c_pad2, "Padded (9:16 Canvas)",
+        c_pad2, "Padded (Canvas)",
         display_bytes=res.get("padded_display_jpeg"),
         download_url=res.get("padded_native_url"),
         filename_stub="padded_backend",
@@ -574,15 +584,14 @@ if res:
         key_prefix=f"job{st.session_state['job_counter']}"
     )
 
-    # Video
-    vurl = res.get("video_url")
     c_video2.markdown('<div class="lv-card-title">AI Cinematic</div>', unsafe_allow_html=True)
+    vurl = res.get("video_url")
     if vurl:
         abs_v = _abs_backend_url(vurl)
         c_video2.markdown(
             f"""
-            <div class="lv-frame" style="display:flex;align-items:center;justify-content:center;">
-              <video autoplay loop muted controls playsinline style="width:100%;height:100%;object-fit:cover;">
+            <div class="lv-frame">
+              <video class="lv-media" autoplay loop muted controls playsinline>
                 <source src="{abs_v}" type="video/mp4">
               </video>
             </div>
@@ -595,7 +604,6 @@ if res:
                                      mime="video/mp4", key=f"job{st.session_state['job_counter']}_video_dl")
     else:
         c_video2.markdown('<div class="lv-frame"></div>', unsafe_allow_html=True)
-        # On-demand video from scene or packshot
         vid_checkbox_key = f"video_toggle_job{st.session_state['job_counter']}"
         if st.checkbox("Generate Video From Scene", value=False, key=vid_checkbox_key):
             src_for_video = res.get("scene_native_url") or res.get("packshot_native_url")
@@ -608,7 +616,7 @@ if res:
                             src_for_video,
                             duration=st.session_state.get("duration_sel", VIDEO_DUR_OPTIONS[0]),
                             resolution=st.session_state.get("resolution_sel", VIDEO_RES_OPTIONS[1]),
-                            preview_aspect=st.session_state.get("preview_aspect","9:16"),
+                            preview_aspect=st.session_state.get("preview_aspect","3:4"),
                             prompt=res.get("scene_prompt_text")
                         )
                     except requests.HTTPError as e: